Uploaded image for project: 'Documentation'
  1. Documentation
  2. DOCUMENT-1155

Update hardware recommendations



    • Type: Improvement
    • Status: Closed
    • Priority: Normal
    • Resolution: Incomplete
    • Component/s: None
    • Labels:
    • Template:
    • QA Risk Assessment:
      Needs Assessment


      As a customer I want to have accurate hardware requirements for my architecture & node count of Puppet Enterprise. Right now, some of the recommendations are too low.

      The main issue is that we specify too low specs for Standard installations, which don't take into account that multiple processes are battling for resources on the same box.
      Another, smaller, issue is that on virtualized nodes a vCPU typically only equates to about 55% of a "real" CPU core if hyperthreading is enabled. This can cause our recommended specs to be off by almost 50%.

      I would recommend the following new spec tables (for CPU/RAM, storage looks OK):

      When using virtual CPUs on a hypervisor that is running on hardware with hyper-threading enabled, add 40% extra vCPUs to the physical CPU core count recommendations below.

      Standard monolithic installs:

      • Trail use (up to 10) => 4 CPU, 8GB RAM
        * Up to 100 => 6 CPU, 10GB RAM
        * Up to 500 => 8 CPU, 12GB RAM
        * Up to 1000 => 10 CPU, 16GB RAM
        * Up to 2500 => 12 CPU, 24GB RAM
        * Above 2500 => switch to large install architecture

      Large installs:

      • 2500 to 20.000 => 16 CPU, 32GB RAM for the PE Primary
      • each compiler (1500-3000 nodes) => 6 CPU, 12GB RAM

      Extra Large installs:

      • 20.000+ => 16 CPU, 32GB RAM for the PE Primary
      • each compiler (1500-3000 nodes) => 6 CPU, 12GB RAM
      • PE-PostgreSQL node => 16 CPU, 128GB RAM




            Unassigned Unassigned
            kevin.reeuwijk Kevin Reeuwijk
            0 Vote for this issue
            7 Start watching this issue



                Zendesk Support